DL-introduction

Assignments for the course Deep Learning & Artificial Neural Networks, MSc of Bioinformatics, KU Leuven, 2023

Part 1 - Supervised learning & generalisation

• Comparing different training algorithms for both clean and noisy data
• Bayesian regularisation for overparametrised networks
• Approximating a 2D surface using a network architecture and tuning procedure by choice

Part 2 - Recurrent neural networks

• Assessing Hopfield networks with 2 and 3 neurons
• Assessing a Hopfield network for digit classification
• Designing and tuning MLP and LSTM networks for time-series prediction

Part 3 - Deep feature learning

• Effect of the number of dimensions for the reconstruction of an image after a PCA compression
• Tuning a stacked autoencoder network
• Dimension reasoning in AlexNet
• Trying some CNN architecture for hand-written digit classification

Part 4 - Generative models

• Comparison of restricted and deep Boltzmann machines for digit image generation
• Examining the behaviour of a DCGAN and WGANs for a class of the CIFAR dataset
• Optimal transport for colour swapping in images